fusion

Tag

Cards List
#fusion

CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs

Hacker News Top · 2026-05-22 Cached

Introduces CODA, a GPU kernel abstraction that expresses Transformer operations as GEMM-plus-epilogue programs to reduce data movement, covering nearly all non-attention computation in a Transformer block.

0 favorites 0 likes
← Previous
← Back to home

Submit Feedback